Theory one · STEM employment
The "geek gene" hypothesis
In 1997, Cambridge psychologist Simon Baron-Cohen found something odd: fathers and grandfathers of autistic children were about twice as likely to work in engineering as fathers of other children.
He built this into a theory — "hyper-systemizing," later "assortative mating" — that people drawn to highly systematic fields like engineering and computing carry genetic traits linked to autism, and that when two such people become parents, those traits combine. The theory made a checkable prediction: tech-heavy regions should show more diagnosed autism. In the 2010s, Baron-Cohen's team tested it in Eindhoven, the Dutch "Silicon Valley," where roughly 30% of jobs are in the IT sector, comparing it with two similar Dutch cities with far less tech industry.
Eindhoven's rate was two to four times higher than the comparison cities — a real, published finding. But Baron-Cohen has also been explicit about where the evidence stops: the popular versions of this story involving Silicon Valley itself, Bangalore, or MIT alumni have never been rigorously studied. Those remain anecdotes, repeated for twenty years, not data.
What survives from the hypothesis
The signal may be real. “STEM” may be the wrong variable.
The geographic story is shaky. The trait-level story has become more interesting.
In a general-population couples study, partners resembled one another on autistic traits, systemizing, and attention-to-detail performance — but not on self-reported empathy. The resemblance was modest and dimension-specific, not a single “autism personality” shared by both partners.
A separate autism-family analysis found spouse Social Responsiveness Scale scores correlated at about r = 0.34. Current autism and intelligence polygenic scores explained little of that phenotypic resemblance.
Parental social-responsiveness traits predicted offspring ASD without using STEM at all: elevated scores in either parent were associated with OR 1.52, and concordantly elevated scores in both parents with OR 1.85.
STEM still matters as an enrichment marker: in a sample of more than 450,000 adults, STEM workers had somewhat higher average autistic-trait scores than non-STEM workers, but the effect was small (Cohen's d ≈ 0.26) and the distributions overlapped heavily.
The strongest clue comes from a Danish registry study of roughly 739,000 children that broke parental occupations into underlying O*NET skill dimensions. Paternal systems/ordering and occupational routineness were positively associated with offspring ASD, while communication and social-skill demands were negatively associated. Once those dimensions were modeled together, non-routine analytical skill did not retain an independent positive association.
That suggests a cleaner model: partly inherited parental traits → occupational sorting, and independently partly inherited parental traits → offspring ASD liability. STEM may enrich some of those traits and shape the mating pool without being the biological exposure itself.
And that changes what a map can test. If the individual-level phenotype is the signal, a tech-heavy county is only a noisy proxy — and its recorded prevalence can still move when the measurement system changes.
The twist
What actually happened in Silicon Valley
California is the one place that can check the popular geographic version of the story, because the state has tracked autism caseloads by county since the early 1990s. The map does not stay put.
Through the 1990s, Santa Clara County — the literal Silicon Valley — saw autism diagnoses among white children double in just seven years, reaching 1.2% by 2000, among the highest rates of any California county. That's the exact pattern that helped inspire the assortative-mating theory in the first place. But researchers who kept following the same data found something the theory never predicted.
After 2000, Santa Clara's rate flattened, then fell — while rates in California's poorest counties kept climbing. By 2013, white children in the state's lowest-income counties had at least double, and in some places up to ten times, the autism rate of white children in Santa Clara and the wealthiest coastal counties. The geography of "who gets more autism" in California didn't stay pinned to the tech industry. It moved to the opposite end of the income scale.
Researchers aren't fully certain why. One live hypothesis: as autism became better known and insurance coverage improved, wealthier families increasingly pursued private diagnosis and treatment — outside the state system this data comes from — making their children statistically disappear from the very count used to track them. If that's right, it isn't that Silicon Valley's kids stopped needing services. It's that the count stopped seeing them — the same lesson as everywhere else in this series, applied to the wealthiest end of the map instead of the poorest.
Theory two · a different population entirely
Not the same immigrants, not the same story
The other geographic pattern people point to involves immigrant communities — and it's worth saying plainly: this is a different population than the tech-sector story, not a variation on it.
The elevated rates researchers have actually documented cluster among refugees and immigrants from specific low-income, low-resource countries of origin — not the highly paid, highly credentialed immigration that fills STEM jobs in tech hubs. The best-studied example is Somali children in Sweden and in Minneapolis, Minnesota.
Open question
Why researchers are still not sure
Unlike the Silicon Valley story, this one doesn't have a tidy access-driven explanation waiting in the wings — because access mostly runs the wrong direction to explain it.
Refugee and low-income immigrant communities generally have less access to insurance, specialists, and private evaluation than the wealthy counties in the last section — the opposite of the conditions that inflated Santa Clara's numbers in the 1990s. Yet the rates are still elevated, and specifically for the more severe, intellectual-disability-linked presentation of autism that's hardest to miss. Researchers have proposed several explanations, and are candid that none is fully confirmed: lower vitamin D from reduced sun exposure at northern latitudes; maternal stress linked to displacement and migration during pregnancy; and a community-mobilization effect, where a cluster of early, visible cases prompted a community to organize, seek evaluations, and find children who might otherwise have gone unlabeled. One 2015 review put it directly: ethnicity and biology alone don't explain the gap.
One pattern is consistent across the research, though: the elevated risk is specifically tied to migration from low-income, low-resource countries of origin, and to migrating near or during pregnancy — not to being an immigrant in general. A software engineer moving from Bangalore or Shanghai to San Jose is not the population these studies describe.
This exact data was, at one point, turned into real harm. In 2008, Somali parents in Minneapolis noticed unusually high numbers of children in autism special-education programs — a legitimate community concern that led to genuine research, including the Stockholm and Minneapolis studies cited above.
It also became a target. Anti-vaccine activists — including Andrew Wakefield, the researcher whose fraudulent, retracted 1998 paper originated the vaccine-autism myth — traveled to Minnesota specifically to tell Somali parents that the MMR vaccine was the cause. MMR vaccination among Somali two-year-olds in Minnesota fell from 92% in 2004 to 42% by 2014. In 2017, Minneapolis had a measles outbreak: 79 confirmed cases, 65 of them in Somali children.
No study has ever found a link between vaccines and autism. The elevated autism numbers in this community were real. The explanation activists gave for them was not — and the harm that false explanation caused was.
The common thread
Two theories that don't actually converge
In neither story does geography turn out to mean quite what it first appears to mean — but it doesn't mean the same other thing in both cases, and resisting the urge to flatten them into one tidy explanation is itself the honest finding.
The Silicon Valley map increasingly looks like a story about which families get seen by which measurement system — a wealth-and-access pattern that has already reversed once and could plausibly reverse again as private care grows outside public tracking. But the trait-level signal underneath the old STEM hypothesis has not disappeared. Couples modestly resemble one another on some autism-related dimensions, parental quantitative traits predict offspring ASD, and occupational decomposition suggests that systems/ordering, routineness, and social demands are more informative than the STEM label or generalized analytical skill. That makes STEM a plausible enrichment marker and mating-pool structure, not the exposure itself.
The immigrant-community story is different and less resolved: access explanations run the wrong direction to fully account for it, which is exactly why it remains a genuine open question rather than a settled one. Both stories, and the false one anti-vaccine activists tried to graft onto the second, share a lesson from this whole series: a map of autism prevalence is rarely just a map of autism. It is also a map of who got looked at, which traits were used as proxies, who entered the measurement system, and who got a name for what they were already seeing in their own child.
The map was never the mechanism. In the tech story, a real trait-level signal may be hiding behind a crude occupational label while the geography is distorted by ascertainment. In the immigrant story, the elevated pattern is different — and still unresolved.